Hi everyone,

I noticed a strange behavior of bitbake, and I am not sure whether it is a bug:

Let say that I have a simple recipe that takes the directory `THISDIR/files/A` and install all the files inside of it:

...
SRC_URI = "file://A/" 
S = "${WORKDIR}/A"
do_install() {
    install -m 644 ${S}/* ${D}
}
...

Let say that I have one file called "my_file" inside of that directory.

It will work great, and I will get a package with "/myfile" -- until I will rename a file to "/myfile1" in the directory. Since the file content stayed the same, do_fetch won't be triggered and as a result, the package will stay the same and have "/myfile".

The only proper way to workaround it was to mark this recipe's do_fetch as nostamp:

do_fetch[nostamp] = "1"

I am currently working with bitbake 1.46.0.

Questions:
1. Is this a known issue?
2. I could not find any reference to a similar issue / a recent change that could have caused the issue, am I doing something wrong here?

Thanks,
Shmuel.

On Wed, 2021-12-22 at 17:54 +0000, Shmuel Hazan wrote:

I noticed a strange behavior of bitbake, and I am not sure whether it is a
bug:

Let say that I have a simple recipe that takes the directory `THISDIR/files/A`
and install all the files inside of it:

...
SRC_URI = "file://A/" 
S = "${WORKDIR}/A"
do_install() {
    install -m 644 ${S}/* ${D}
}
...

Let say that I have one file called "my_file" inside of that directory.

It will work great, and I will get a package with "/myfile" -- until I will
rename a file to "/myfile1" in the directory. Since the file content stayed
the same, do_fetch won't be triggered and as a result, the package will stay
the same and have "/myfile".

The only proper way to workaround it was to mark this recipe's do_fetch as
nostamp:

do_fetch[nostamp] = "1"

I am currently working with bitbake 1.46.0.

Questions:
1. Is this a known issue?
2. I could not find any reference to a similar issue / a recent change that
could have caused the issue, am I doing something wrong here?

I'm pretty sure we fixed bugs like that in more recent versions.

Cheers,

Richard

On Wed, 2021-12-22 at 18:10 +0000, Richard Purdie wrote:

On Wed, 2021-12-22 at 17:54 +0000, Shmuel Hazan wrote:

I noticed a strange behavior of bitbake, and I am not sure whether
it is a
bug:

Let say that I have a simple recipe that takes the directory
`THISDIR/files/A`
and install all the files inside of it:

...
SRC_URI = "file://A/" 
S = "${WORKDIR}/A"
do_install() {
    install -m 644 ${S}/* ${D}
}
...

Let say that I have one file called "my_file" inside of that
directory.

It will work great, and I will get a package with "/myfile" --
until I will
rename a file to "/myfile1" in the directory. Since the file
content stayed
the same, do_fetch won't be triggered and as a result, the package
will stay
the same and have "/myfile".

The only proper way to workaround it was to mark this recipe's
do_fetch as
nostamp:

do_fetch[nostamp] = "1"

I am currently working with bitbake 1.46.0.

Questions:
1. Is this a known issue?
2. I could not find any reference to a similar issue / a recent
change that
could have caused the issue, am I doing something wrong here?

I'm pretty sure we fixed bugs like that in more recent versions.

Thanks! 

For a reference, b4975d2ecf615ac4c240808fbc5a3f879a93846b
(fetch2/checksum/siggen: Fix taskhashes not tracking file directories)
from 2~ months ago seems to solve that issue.

I see that the checksum code was not changed for a long time, is there
a chance that someone would accept a backport of this commit to
Dunfell/1.46?

Cheers,

Richard